The tests I ran included 24 hours of continuous application of common user programs: ls, du, find, mkdir and rm. These programs were invoked from a simple driver shell script that ran each one of them in turn. First I ran the script on an unmounted /usr/local file system. Then I mounted Wrapfs (once) on top of /usr/local, and reran the script. I used the time utility to measure how much system time was consumed by each run.
Preliminary performance measurements showed that interposing the Wrapfs file system once on top of UFS resulted in degradation ranging from 3.5% (using Solaris 2.4 x86 on a P90 with 24MB RAM and an IDE disk) to 6.4% (using Solaris 2.4 SPARC on an SS2 with 64MB RAM and a SCSI disk) in reported ``system'' time.
Therefore, the overhead of the first version of Wrapfs is comparable to the mechanisms implemented by Skinner and Wong [Skinner93] (up to 10%) and the UCLA stackable layers project [Heidemann94] (3%).